Development of Machine Learning-based Shelf Life Prediction Models for Pearl Millet (Pennisetum glaucum L.) Grains using Oxidation Kinetics Data

By observing oxidation kinetics, this study examined how storage temperature (5°C, 25°C, 45°C) and duration (0 to 120 days) affect the quality of raw and fermented pearl millet grains. The study also compared the accuracy of different machine-learning approaches for predicting shelf life of pearl millet grains. The results showed that the levels of free fatty acid (FFA), acid value (AV), peroxide values (PV) and lipase activity (LA) increased with the temperature and duration of storage, regardless of the treatment. For raw grains, the FFA, AV, PV and LA content varied from 0.89% to 6.23%, 1.3 to 8.84 mg NaOH 100g-1, 5 to 88.33 mEq kg-1 of flour, and 4.34 to 23.47 mg KOH g-1, respectively during 120 days of storage over the storage temperature range under consideration. On the other hand, in the case of fermented grains, the values of FFA, AV, PV and LA content ranged from 0.85% to 4.52%, 1.2 to 6.41 mg NaOH 100g-1, 5 to 51.67 mEq kg-1 of flour, and 4.29 to 12.36 mg KOH g-1, respectively. The FFA, AV, and PV values for raw and fermented grains were used to estimate the shelf life using oxidation kinetics data. The kinetics data fit a pseudo- zero-order reaction model better than a first-order reaction model (R2 =0.8901 to 0.9927). Among all the machine learning techniques, artificial neural network (ANN) was found to be a better predictor with the least error functions and higher accuracy (R2 =0.9847 to 0.9969) as compared with the Gradient Boosting and the Support Vector Machine models.